Rental pricing for Nvidia’s older H100 graphics processing units has strengthened, rising about 22% over the past month, according to the Ornn H100 SXM Index. The move is drawing attention from investors as Nvidia prepares to launch its next-generation Rubin platform, with the company positioning newer chips alongside continued demand for earlier systems.
The data suggests customers are still willing to pay for H100 capacity even as more advanced options emerge, complicating the bearish view that the AI chip market will rapidly move past Nvidia’s installed base.
Key takeaways
- Price move: H100 GPU rental costs rose roughly 22% month over month, reaching $3.28 per hour in the index.
- Catalyst: The market is pricing in continued demand for the H100 despite Nvidia’s upcoming Rubin platform, which would make the H100 two generations old.
- Key implication: The “installed-base depreciation” thesis appears less certain as older chips hold rental value.
- Competitive angle: Multi-tier usage may persist, with premium customers paying for newest chips while cost-sensitive customers rent older models like the A100 and H100.
What drove the H100 rental rebound
The rebound centers on live cloud GPU rental pricing for the H100, measured through the Ornn H100 SXM Index. On X, Nvidia CEO Jensen Huang described Nvidia compute as “fungible, durable and highly rentable,” while Ornn’s exchange cited the index’s monthly increase and referenced the H100’s age as a key part of the surprise: the chip is described as a three-year-old training component.
According to the Ornn index, the H100’s rental price climbed to $3.28 an hour and is up 22% versus a month ago. The underlying message to investors is straightforward: if rental pricing for a now-older training chip can rise as new generations approach, then demand may remain anchored in ongoing training workloads and enterprise-scale AI deployments that do not instantaneously replace older infrastructure.
Why this matters even as Rubin rolls out
Nvidia’s Rubin platform—and specifically the R100 variant cited in the article—would make the H100 two generations old. The expectation in many bull–bear debates is that as newer architectures arrive, utilization and purchasing shift quickly toward the latest models, pressuring older-chip prices.
Instead, the rental market as reflected in the index shows the opposite trend in the near term. While the H100 (and the prior generation A100) has seen pricing come down from peaks during the early phase of the generative AI boom, the recent increase indicates that older GPUs are still finding buyers and renters—likely because organizations balance performance targets, deployment timing, and budget constraints.
Market reaction: durability across generations
The broader implication for Nvidia is that the company may benefit from a layered product demand curve. The article points to an environment where customers with different cost and performance priorities pay different prices for different generations rather than moving in lockstep to the newest chip.
It also highlights that demand is not limited to the H100. Even the A100 remains in use, with at least some customers preferring its lower per-hour rental cost. In that framework, Nvidia’s business model can support continued revenue and volume from both the newest chips and the older generations that remain “good enough” for specific training and deployment needs.
What skeptics have warned—and what the new pricing suggests
Some investors have argued that depreciation risk could weigh on Nvidia’s economics and on the sustainability of the broader AI buildout. The article attributes that view to a theory associated with Michael Burry, known for “The Big Short,” which centers on the idea that hyperscalers and other cloud players may face financial pressure if chip values fall faster than expected. If GPUs lose value quickly, it can affect both resale expectations and the economic rationale for continued capacity expansion, depending on how services are priced relative to compute costs.
However, the article says there is evidence that Nvidia GPUs retain value better than skeptics anticipated. In the current snapshot, the H100’s rental price increase—despite the passage of time and the approach of newer architectures—runs counter to the most pessimistic interpretation of fast depreciation.
Bigger picture for investors
For investors monitoring Nvidia, the H100 pricing durability provides a tangible read-through to real-world demand. While semiconductor demand remains cyclical and competitive pressures from custom accelerators built by large cloud customers are widely discussed, the rental market signal suggests that Nvidia’s older chips are still economically viable in the cloud.
The data also offers a reason to reassess how quickly the market expects the AI spending cycle to shift away from the installed base. If rental prices can rise into the next generation transition, then replacement cycles may be slower, more staggered, or more segmented than many bearish scenarios assume.
What to watch next: Investors are likely to focus on Nvidia’s guidance for demand across chip generations, updates on Rubin’s adoption timeline, and whether pricing power persists beyond near-term rental moves. Upcoming milestones to monitor include Nvidia’s next earnings cycle and any additional commentary on customer deployment plans for newer versus legacy GPUs.
